Can image-to-3D make realistic 3D-printable human figurines without losing facial identity?
LoudPoem9492 · reddit · 2026-07-22
The poster is trying to make highly realistic 3D-printable figurines of real people from photos, and says the main failure mode across Hunyuan 3D, Hyper3D Rodin, Hi3D, Tripo, Pixal3D, ComfyUI workflows, and several open-source tools is the same: faces lose identity and turn generic.
They’ve already tried multiple preprocessing tricks — transparent/white backgrounds, upscaling, cleanup, isolated subjects, different prompts and resolutions — but likeness is still not good enough. They’re asking whether anyone has gotten better results with workflows such as:
- multi-view generation before reconstruction
- Flux-based pipelines
- local Hunyuan or TRELLIS
- Gaussian Splatting / NeRF
- photogrammetry plus AI assistance
- Blender or academic hybrid workflows
Their core question: have we hit the current ceiling of image-to-3D for realistic human figurines, or is there still a noticeably better path for facial resemblance?
More from Multimodal
- K3 turns a 3D model and one prompt into a working scene in 30 minutes — teortaxesTex · 2026-07-22
- A Reddit user shares a ComfyUI stack for Krea 2 and image upscaling — darlens13 · 2026-07-22
- Appearance Pointers add localized multimodal control to diffusion transformers — Rahul Sajnani · 2026-07-22
- A liminal-spaces AI art experiment is now available on Mac and Windows — AIandDesign · 2026-07-22
- Custom LoRA turns family photos into a Spider-Gwen multimodal demo — alecubudulecu · 2026-07-22
- Claude Code and Pika MCP turn a rough image into a full character sheet — ytjessie_ · 2026-07-22